14-18 apr 08 adm-aeolus wind retrievals for nwp – 9th

14-18 Apr 08 ADM-Aeolus Wind Retrievals for NWP – 9th IWW Annapolis Slide 1

ADM-Aeoluswind retrievals for NWP

DGH TanECMWF

Acknowledgments:ESA (Mission Experts Division & Aeolus project team)Aeolus Mission Advisory GroupLevel-1B/2A/2B Development Teams


ECMWF preparations 2002-2004Example for other NWP centres

Level-2B processor developmentECMWF is lead institute, 5 sub-contractors

2004-present

Other ongoing work/operational phaseGSOV, Cal/Val, In-orbit commissioning

ECMWF to generate operational L2B/L2C products, monitor & assimilate Aeolus data, assess impact on NWP

Maintain, develop & distribute L2B processorOn behalf of ESA, using NWP-SAF approach

Overview – almost ready for launch


Product for assimilation – L2B 50km hlos windADM-Aeolus Baseline UV lidar (355 nm) with two receivers

- Mie (aerosol), Rayleigh (molecules)

- both use direct detection

Wind profiles from surface to 30 km with resolution varying from 0.5 to 2 km

- vertical bins configurable in flight

- HLOS component only

- direction 7º from zonal at equator

- 6 hour coverage shown

Burst mode 1 obs ~ 50 measScalable system

L2C – adds along-track wind component after data assimilation


90% of Rayleigh data have accuracy better than 2 m/s

In priority areas (filling data gaps in tropics & over oceans)

Complemented by good Mie data from cloud-tops/cirrus (5 to 10%)

Tan & AnderssonQJRMS 2005

60°SS

S 30°S

0°0°

N 30°N

60°NN

150°W

150°W 120°W

120°W 90°W

90°W 60°W

60°W 30°W

30°W 0°

0° 30°E

30°E 60°E

60°E 90°E

90°E 120°E

120°E 150°E

150°E

y g

0.5

0.6

0.7

0.8

0.9

1.0

1%

25%

50%

75%

100%

60°SS

S 30°S

0°0°

N 30°N

60°NN

150°W

150°W 120°W

120°W 90°W

90°W 60°W

60°W 30°W

30°W 0°

0° 30°E

30°E 60°E

60°E 90°E

90°E 120°E

120°E 150°E

150°E

0.5

0.6

0.7

0.8

0.9

1.0

1%

25%

50%

75%

100%

LIPAS-simulated HLOS data – operational processors later

L2B data simulated using ECMWF clouds …Yield (data meeting mission requirements in % terms) at 10 km


S.Hem

0.0 0.5 1.0 1.51000

100

… & impact studied via assimilation ensembles

Spread in zonal wind (U, m/s)Scaling factor ~ 2 for wind

error

Tropics, N. & S. Hem all similar

Simulated DWL adds value at all altitudes and in longer-range forecasts (T+48,T+120)

Differences significant (T-test)Supported by information

content diagnosticsCheaper than OSSEs

ADM-Aeolus

NoSondes

Pres

sure

(hPa

)

Zonal wind (m/s)

p<0.001

12-hr fc impact (Tan et al QJRMS 2007)


Global information content - consistentMike Fisher for Entropy Reduction & DFS

S ~ log( det( PA ) )

~ tr ( log ( J’’ -1 ) )

J’’ = 4d-var Hessian

PA = analysis error covar.

DWL data are accurate and fill data gaps

subject to usual caveats about simulated data

5027890688N_Obs

27433707Deg_Free_Sig

31234830Entropy_Reduction

(“Info bits”)

2 ⎯ 3 %Redundancy

18.324.5N_Obs/Deg_Free_Sig

0.0620.053Info bits per obs

HLOSu,v to 55 hPaData considered

Simulated DWLTEMP/PILOT


Observation Processing

Data Flow at ECMWF

Non-IFS processing

Observation Screening

Assimilation Algorithm

Diagnostic post-processing

“Bufr2ODB”Convert BUFR to ODB format

Recognize HLOS as new known observable

IFS “Screening Job”Check completeness of report, blacklisting

Background Quality Control

IFS “4D-VAR”Implement HLOS in FWD, TL & ADJ Codes

Variational Quality Control

“Obstat” etc (Lars Isaksen)Recognize HLOS for statistics

Rms, bias, histograms

Prototype Level-2B (LIPAS simulation, includes representativeness error)

Analysis

Assimilation of prototype ADM-Aeolus data2003/4: introduced L2B hlos as new observed quantity in 4d-Var


Observation Processing

Data Flow at ECMWF

Non-IFS processing

Observation Screening

Assimilation Algorithm

Diagnostic post-processing

“Bufr2ODB”Convert BUFR to ODB format

Recognize HLOS as new known observable

IFS “Screening Job”Check completeness of report, blacklisting

Background Quality Control

IFS “4D-VAR”Implement HLOS in FWD, TL & ADJ Codes

Variational Quality Control

“Obstat” etc (Lars Isaksen)Recognize HLOS for statistics

Rms, bias, histograms

Level-1B data

(67 1-km measurements)

Analysis

Assimilation of prototype ADM-Aeolus data2004-: Receive L1B data & L2B processing at NWP centres

L2BP (1 50-km observation)


Aeolus Ground Segment & Data Flows - schematic view

Level-2B processor will run in different environmentsECMWF will supply source code - use as standalone or callable subroutine


Retrievals account for receiver properties …

Tan et al Tellus A 2008

Dabas et al same issueMie light reflected

into Rayleigh channelRayleigh wind

algorithm includes correction term involving scattering ratio (s)

ADM-Aeolus Optical Receiver - Astrium Satellites


… and for atmospheric scattering properties

Line shapeDepends on P, T and s

Rayleigh(P,T)

Mie

Transmissionthrough the Double FP

Light transmitted through TA and TB Response (function of fD)

Response curveFunction of P, T and s

ILIAD – Impact of P & T and backscatter ratio on Rayleigh Responses - Dabas Meteo-France, Flamant IPSL

1km-scale spectra are selectively averagedAccount for atmospheric variability - improve SNR


Retrievals validated for idealized broken multi-layer clouds – E2S simulator + operational processing chain

Specified cloud layers

Retrieved clouds and aerosol

Retrieved Rayleigh winds are accurate in non-cloudy air

50 km

Retrieved Mie winds are accurate in cloud and aerosol layers

Specified wind=50 m/s


Real scattering measurements obtained from the LITE mission

ESA’s software (E2S) is used to simulate what ADM-Aeolus would ‘see’

The L1B software retrieves scattering ratio at the 1 km measurement resolution

Our input not perfect

Realistic scenes simulated


-100 -80 -60 -40 -20 0 20 40 60 80 100

0

5

10

15

20

25

HLOS wind [m/s]A

ltitu

de [k

m]

E2S input (238)2b Rayleigh Cloudy (11)2b Rayleigh Clear (14)1b Rayleigh obs (24)2b Mie Cloudy (12)2b Mie Clear (14)1b Mie obs (8)

Wind retrieval validated in the presence of heterogeneous clouds and wind – E2S simulation

Retrieved Mie winds revealed systematic error in L1B input

Level-1B

Ray

leig

h m

olec

ular

Mie

pa

rtic

les

Retrieved Rayleigh winds are accurate – being refined


Summary – Day-1 system on track1. Level-2B hlos winds – primary product for assimilation

a. Account for more effects than L1B products

b. Will be generated in several environments

c. Motivated strategy to distribute source code

2. Main algorithm components developed & validated

a. Release 1.33 available – development/beta-testing

b. Documentation and Installation Tests

c. Portable – tested on several Linux platforms

3. Ongoing scientific and technical development

a. Sensitivity to inputs, QC/screening, weighting options

4. Contact points – ESA and/or ECMWF


Level-2B hlos error estimates – reqts met

Poli/Dabas

Meteo-France


ADM-Aeolus addresses key observational needs

Objectives, wind observation requirements, DWL instrument, viewing geometry

Implementation well-advanced for launch in 2009

Space and ground segments

HLOS wind product (L2B data, algorithm, portable s-ware)

Cloud and aerosol products (L2A data)

Experimental campaigns and calibration/validation

Studies with wind lidar data support theoretical expectations

Data simulations, NWP data impact studies (assimilation ensembles as alternative to OSSEs, + information content)

Airborne DWL (Weissman). Tropical assimilation (Zagar).

Overview – why expectations are so high


Baker et al 1995, BAMS

ESA 1999 Report for Assessment (Stoffelen et al 2005, BAMS) and 2007/8 Science Report

Weissman and Cardinali 2006, QJRMS

N. Zagar & co-authors, QJRMS & Tellus A

Tan & Andersson 2005, QJRMS

Tan et al 2007, QJRMS

Tan et al 2008, Tellus A (Special Issue on ADM-Aeolus)

Key references


Background for ADM-AeolusMeasurement Concept

• Backscatter signal

• Aeolus winds are derived from Doppler shift of aerosols andmolecules along lidar line-of-sight

• Error estimates, cloud & aerosol properties derived from signal strength

AquaCALIPSOCloudSat

PARASOL

Aura

CALIPSO lidar – vertical cross sections of backscatter


Background for ADM-AeolusMeasurement Concept

• Backscatter signal

• Winds are derived from Doppler shift of aerosols andmolecules along lidar line-of-sight

• Error estimates, cloud & aerosol properties derived from signal strength


ADM-Aeolus Space Segment – preparation/testing of1) structural-thermal model 2) lidar transmitter/receiver


5.1 Prototype Level-2C ProcessingIngestion of L1B.bufr into the assimilation system

L1B obs locations within ODB (internal Observation DataBase)

Assimilation of HLOS observations (L1B/L2B)

Corresponding analysis increments (Z100)

-0.5 0.51480

14801480

1520 1520 1520

1560 15601560

16001600

1600

1600

1600

1640 1640 1640

16401640

60°S

30°S

0°

30°N

60°N

120°W

120°W 60°W

60°W 0°

0° 60°E

60°E 120°E

120°E

p

-2.5-2-1.5-1-0.5-0.25-0.10.10.250.511.522.5


5.2 Key assimilation operatorsTan 2008 ECMWF Seminar Proceedings

HLOS, TL and AD

H = - u sin φ - v cos φ

dH = - du sin φ - dv cos φ

dH* = ( - dy sin φ, - dy cos φ )T

Generalize to layer averages later

Background error

Same as for u and v (assuming isotropy)

Persistence and/or representativeness error

Prototype quality control

Adapt local practice for u and v


2a-4. Other NWP configurations

L2C Product(optional)

Assimilation

(L2C Processor)

Other observations Forecast ModelL2B Product

L2B Processor

L1B ProductProcessingParametersAuxiliary Data

Meteorologicalproducts

Integrated or Standalone, other options possible.

Possibility to modify algorithm source code.


1a/b. What are Level-2B/2C Products?

L2C Product

Assimilation

(L2C Processor)


L2B Processor


Retrieval algorithm accounts for T & p effects in Rayleigh channel signal

Meteorologically representative wind vector profiles

Meteorologically representative hlos profiles

Measurements grouped and weighted according to features detected in atmospheric scene (primarily clouds & aerosol)



1a/b. What are Level-2B/2C Products?2B: Meteorologically representative HLOS profiles

─ retrieval algs applied to Level-1B data, 2B-output suitable as input to data assimilation─ auxiliary input data: T & p, Rayleigh-Brillouin

response data, etc2C: Meteorologically representative wind vector profiles

─ result of a data assimilation algorithm, combining Level-2B with other data/weather forecast model

How do they differ from Level-1B Products?

─ Rayleigh channel retrieval accounts for T & p effects

─ measurements grouped/weighted by features detected in the atmospheric scene (primarily clouds & aerosol)


2a. Who will make Level-2B/2C Products?ECMWF for “operational” Level-2B/2C products─ Processing integrated with data assimilation system─ Products in ESA’s Earth Explorer file format

available from ESA (Long-Term Archive)ESA LTA for Level-2B late- & re-processing─ Level-1B missing ECMWF’s operational schedule─ New processing parameters/auxiliary inputsOther Numerical Weather Prediction centres─ Different operational schedule/assimilation strategy─ Different processing params/aux inputs/algorithmsResearch institutes & general scientific users─ Different processing params/aux inputs/algorithms


2a-1. ECMWF “operational” configuration

L2C Product

Assimilation

(L2C Processor)


L2B Processor



Integrated with assimilation system


2a-2. ESA-LTA late- and re-processing

L2B Product

L2B Processor


Standalone configuration


2a-3. Research/general scientific use


L2B Processor


Standalone configuration.Possibility to modify algorithm source code.


2b. Why distribute L2BP Source Code?Distribution of executable binaries only permits

─ limited number of computing platforms

─ different settings in processing parameters input file─ thresholds for QC, cloud detection

─ different auxiliary inputs─ option to use own meteorological data (T & p) in

place of ECMWF aux met data (available from LTA)Provide maximum flexibility for other centres/institutes to generate their own products

─ different operational schedule/assimilation strategy

─ scope to improve algorithms─ feed into new releases of the operational processor


3a. How it works – Tan et al Tellus A 2008Rayleigh channel HLOS retrieval – Dabas et al, Tellus A

─ R = (A-B) / (A+B) and HLOS = F-1 (R;T,p,s)

─ T and p are auxiliary inputs

─ correction for Mie contamination, using estimate of scattering ratio s

Mie channel HLOS retrieval

─ peak-finding algorithm (4-parameter fit as per L1B)

Retrieval inputs are scene-weighted

─ ACCD = Σ ACCDm Wm, Wm between 0 and 1

Error estimate provided for every Rayleigh & Mie hlos

─ dominant contributions are SNR in each channel


4. Distribution of L2BP softwareSoftware releases issued by ECMWF/ESA

─ Details & timings to be determined

─ Probably via registration with ECMWF and/or ESA

─ Source code and scripts for installation─ Fortran90, some C support─ Developed/tested under several compilers

─ Suite of unit tests with expected test output

─ Documentation─ Software Release Note─ Software Users’ Manual─ Definitions of file formats (IODD), ATBD, etc.


ConclusionsExpectations for ADM-Aeolus are high

─ On track for producing major benefits in NWP─ Meeting the mission requirements for vertical

resolution & accuracy─ Extending to stratosphere, re-analysis─ Our software available to NWP/science community

─ Combine with other observations─ Height assignment for AMVs─ Complement other cloud/aerosol missions

─ Related research─ Background error specification


L1B

ODB

4DVar

5.3 L2BP integration within an assimilation system

screening

Background T p

L1B-odb lat lon

AMD-odb

Background hlos

L2B-odb

Obs - Bg, BgQC, etc

L2B Processor

L1B% AMD%

L2B%

Wrapper module copies “odb”

variables to & from

% data structures

(hidden from

screening)

C.f. RTTOV +

SSMITCWV


L2C

AMD

L2B

L1B

5.4 Overview data flow – standalone mode

EE L2BP_standalone

L2B Processor

L1B% AMD%

L2B%


5.5 Principal Guidance to Met Centres1. How to install and test the standalone version

Source code, documentation, unix scripts and test data (EE format) supplied

Useful tool for inter-comparison purposes

2. Interface requirements for integrated-assimilation mode

Generation of auxiliary meteorological data

Wrapper module between “odb” and L2B processor used as a callable subroutine within assimilation.x

Both to occur during Screening

Facilitates assimilation of Aeolus data

Assimilation outputs at discretion of each met centre

14-18 apr 08 adm-aeolus wind retrievals for nwp – 9th

Documents